SISA papers

Estimating the variance of a proportion when the data are clustered. Comparing the variance correction factor and jackknive resampling technique.

Comparison of the SISA spreadsheet to estimate the variance of a proportion using a clustered sampling design with the jacknive method as used in Wesvar. Basically the two methods produce similar results. However, when the data is skewed both methods become unreliable particularly when the intracorrelation is also low.

On the Number Needed to Treat.

In this paper the use of the NNT is compared with other measures of association. The NNT has the advantage of being easy to interpret particularly in relation to cost benefit analysis. However, the measure has a number of very undesirable properties such as that the range is interrupted, there are no valid values between -1 and 1, and that the confidence interval has to be able to incorporate infinity. These problems are particularly important if one wants to use the NNT for statistical significance testing.

On the mathematical relationship between the number of events in which people are injured and the number of people injured.

This previously published paper explains the Poisson distribution. In the paper the distribution is used to make a connection between the number of persons who experience an event and the probability that the event (for example an accident) happens to the same person once, twice or more often

Life expectancy and SMR applied to migrant groups living in Amsterdam, The Netherlands.

Paper which studies the apparent contradiction between life table data which shows that migrants living in Amsterdam have a high life expectancy, and the SMR which shows that they also have a high mortality. The study shows that the SMR is very much influenced by differences in population age structure between groups studied. Although the SMR is meant to correct for such differences, the SMR is highly unreliable if populations are too different.

Method of small p-values

Short discussion of the method which is used to estimate exact double sided p-values, such as for the binomial, poisson and Fisher test.

Discounting and mortality adjusting Years of Potential Life Lost (YPLL)

Years of Potential Life Lost (YPLL) or Potential Years of Life Lost (PYLL) is an often used statistic in practical epidemiology and demography. The measure is easy to calculate, easy to understand and has a strong intuitive appeal. The YPLL measures for a group of individuals the total number of years these people would have additionally lived up to some point in the future, would they not have died from a particular cause of death. Mostly, as the age where life is "lost" by a premature death, the life expectancy for a population or the age of productivity, from 15 to 65 or 70, are chosen. Besides the obvious advantages of the YPLL there are a number of problems. Two of these problems are discussed in this paper, one the practical problem of correction of the YPLL for people surviving but then dying from an other cause of death, second the theoretical and philosophical problem of discounting of the value of life lived in the far away future.

Calculating the discounted YPLL - annotated.

Annotated version of part of the above paper.

Note on discounting benefits.

This footnote is about the formula on discounting benefits such as in calculating the discounted YPLL, or to discount health care costs and benefits.

Design, data weighing and designeffects in Dutch regional health surveys

In stratified sampling designs the data mostly has to be weighted to report on the population level. This introduces the designeffect, which will lower the reliability of reported statistics. In this article the calculation of the designeffect is discussed and demonstrated. After that the designeffect is calculated for a number of health surveys. The designeffects observed ranged from 1.00, in case of the self weighing design, to 1.85, in a design based on same size samples drawn from very differently sized population groups. The designeffect can be important, particularly in the analysis of smaller samples. Considering designeffects in sample size calculations previous to collecting data is therefore important.

Use of Bonferroni Multiple Testing Correction With an Internet Based Calculator. An Analysis of User Behaviour.

There is concern about the application of the Bonferroni correction in research. This paper studies how the procedure is used by practitioners on an internet based calculator. Each requests for a Bonferroni correction is logged into a single line of data. The data studied concerns the year 2018. After removal of invalid requests the data concerns 9682 lines of data pertaining to 3624 different IP addresses. Most of the users do only one Bonferroni request, there is a strong preference for a p-value of 0.05. Of the requests 16.4% specified more than 25 p-values as the denominator for the Bonferroni correction. 16.6% of users specified a correlation whereby 44.3% of the correlations was 0.5 or larger. Around 15% of users requested a Holmes correction. Particularly the high number of multiple tests led to a large proportion of Bonferroni adjusted p-values being low. Correlation correction was applied not very often, however, when correlation correction was requested the specified correlation was relatively large and had a significant impact on the result of the Bonferroni correction. Holms correction is not often requested. A number of recommendations are done for the use of Bonferroni correction.

Survey data weighing questions and answers

Answers a number of questions about survey data weighing

TOP of page

Compare Car Rentals!
Help SISA and compare two rental cars!
An easy way to find the best option.

SISA Research Papers